Signature terms
Canonical definitions for the three terms introduced in Beyond Recall (Gulaya 2026).
Representational accuracy
A measure of how faithfully an AI system captures a specific person’s interpretation. Distinct from recall, persona consistency, and preference matching. Introduced in Beyond Recall (Gulaya 2026) as a distinct evaluation axis for human-AI alignment.
Behavioral Specification
A structured interpretive-layer artifact that operationalizes how a specific person reasons, decides, and communicates. Served as context to a language model above retrieval rather than as a replacement for it. Aggressively compresses a person’s data into interpretive patterns. Reference implementation in Beyond Recall (Gulaya 2026).
Beyond Recall
Title of the preprint by Aarik Gulaya (2026) introducing representational accuracy as a distinct evaluation axis and the Behavioral Specification as the interpretive layer that produces it. Evaluated across 14 public-domain autobiographical corpora using a calibrated 5-judge LLM panel against four commercial memory systems. Read the preprint.